What is normal for your server at this hour

This article is for anyone who sees a CPU, memory or other usage number on a server and wants to know whether it's actually high — or just normal for that time of day.

#Overview

• A number on its own doesn't say much. 70% CPU at 02:00, while the nightly backup runs, can be completely normal. The same 70% at 14:00 on a quiet afternoon is not.
• InfraNest learns what is normal for each of your servers, hour by hour through the week, and shows it next to your readings.
• You need the agent on the server, or a cloud provider that reports metrics, and the server needs four weeks of readings before its normal range appears.

#Find your server's normal range

1. Open the server and go to the Metrics tab. Each graph has a shaded band behind the line showing what's normal for this hour of the week.
2. Hover over a graph to read the range at that moment (for example, "normal 10–25%").
3. Use Normal range above the graphs to turn the shaded band on or off.
4. Check the server's Dashboard view too — the small graphs at the top show the same shaded band, with a line underneath reading something like "normal now 10–25%".

#Set an alert rule using the normal range

1. Go to Alert rules and click Create alert rule.
2. As you choose the metric, look under the last 30 days section — it tells you how high the server normally goes and at which hour of the week.
3. Pick a threshold above that normal range so the alert only fires when something is genuinely unusual.
4. Click Save to finish.

#Review an unusual hour

When a reading is clearly above its normal range for that hour of the week, a message shows at the top of the server's page, for example:

"CPU is unusual for this hour of the week. 68% on Tuesday 14:00. At this hour of the week it is normally between 10% and 25%."

1. Click See the graph to look at the reading next to the other graphs, with deploys, restarts and changes marked on them.
2. If you know why it happened (a planned import, a one-off job), click This was expected. The message goes away.

If the same reading was unusual at different times of the day, each occurrence gets its own message, so you can answer for each one separately. Hours inside a maintenance window that pauses your alerts are not flagged.

NoteThese messages only appear on the server's own page. They don't send an email, a chat message or a notification, and they don't appear on Home. Your alert rules keep working exactly as before.

#Before the server has four weeks of its own

A new server has no history yet, so nothing is shown at first, and the Metrics tab tells you how many weeks of data it has collected so far.

If the new server has a similar one already running (behind the same load balancer, or doing the same job, such as another web server), InfraNest borrows that server's normal range in the meantime. You'll see something like "borrowed from web-02" next to the range, until the new server builds up four weeks of its own.

#Tips

• Each hour of the week is learned on its own — Monday 02:00 is compared with earlier Mondays at 02:00, not with the rest of the day.
• Outages and maintenance windows are left out of the learning, so a bad night doesn't become "normal".
• A memory leak that climbs steadily until every restart isn't learned as normal either.
• The normal range starts learning again from scratch after a resize, a reinstall, or when the server starts doing a different job.

TipInfraNest looks back up to twelve weeks when learning what's normal, even if your plan keeps less history in the graphs. Times are always shown on your own clock, so the band moves with your server's real schedule.